The work performed has been perfectly aligned with the goals of my ERC project. In particular, I have significantly advanced fundamental understanding of multi-agent decision making under uncertainty, developed novel scalable and provably safe algorithms for these systems and have verified the methods on realistic power system simulation testbeds in collaborations with power system researchers. The work can be discussed in two main categories.
On the electricity market aspect, the novelty in our work was to derive new market mechanisms, motivated by advances in auction theory from the fields of computer science, game theory and economics. These mechanisms clarified the potential of jeopardizing the current electricity markets through collusion and shill bidding and offered alternatives to address these shortcomings. Furthermore, using the tools from coalition game theory, we developed a novel approach to coordinate the reserve markets across Europe. This will significantly enhance the stability of the power system and improve its efficiency. The reason is that the increased use of renewable sources of energy will require an increased need for reserve markets, which help in balance of supply and demand of electricity in real-time. We quantified the benefits of pooling together all European reserve markets both in terms of grid stability and in terms of reduction in cost of achieving the grid stability.
On distributed control algorithms for large-scale systems, we have also achieved two main novel methodologies.
First, in the case in which the subsystems in the large-scale system share a global objective (for example, stability in an interconnected power system), we developed a new approach to distributed control, which would ensure the optimality of the control laws derived, despite the non-convexity of the problem. The approach was based on a novel insight we developed that we can come up with various parameterizations of a controller, some of which lend themselves to tractable distributed design approaches. Furthermore, we developed a novel data-driven approach to learning distributed control problems, motivated by advances in black-box learning and optimization. This approach was not proposed in past works, and at the same time is highly relevant. The reason is that we are increasingly controlling complex systems, which are hard to model. And at the same time, we are increasingly probing our systems and obtaining data from them at an unprecedented way.
Second, in terms of systems that might have individual objectives and not share a single global objective, we developed game theoretic algorithms to learn optimal local decision rules. These algorithms enable the subsystems to learn the optimal actions only using local information and hence, do not jeopardize privacy of the subsystems. We showed the applicability of our result to several case studies and examples arising in real-time electricity markets.
